Previously, interband cascade light-emitting diodes (ICLEDs) grown on GaSb substrates have been demonstrated as useful emitters in the mid-wave infrared (MWIR) region of 3 – 5 µm for room-temperature (RT) continuous wave (CW) operation [1,2]. Transferring this technology to growth on Silicon substrates would be advantageous for applications in chemical sensing and IR scene projectors (IRSPs), providing improved manufacturability through direct integration onto these circuits. This presentation will discuss the comparison of high-performance ICLEDs grown at NRL on GaSb/Si buffers that were grown at UNM and on lattice-matched GaSb substrates, including L-I characteristics, cross-section transmission electron microscopy (XTEM) and x-ray reciprocal space mapping (RSM).

The growth of GaSb/Si involves GaSb buffer layers which were grown on Silicon (001) with a 4° offcut towards (111). The native oxide was removed using a dilute HF solution to obtain a hydrogen-passivated surface. To achieve III-V nucleation on Silicon, a ~10 nm thick AlSb layer was grown at a substrate temperature of 500°C followed by a 1 µm buffer layer and an antimony cap to prevent oxidation. The GaSb/Si wafers were then transferred to NRL where an additional 2-3 µm GaSb buffer and the ungrouped active ICLED stages were grown. This same 22-stage structure was grown on a GaSb substrate as a control sample. Accounting for differences in architecture, the ICLED structures grown on Silicon show efficiencies that are 75% of those measured in ICLEDs grown on GaSb. At 100 mA, 200-µm-diameter mesas produce 184 µW CW at 25°C and 140 µW at 85°C.

Threading dislocations were observed in GaSb buffer grown on Si from the XTEM images, showing a higher density near the Silicon substrate but reduced near the ICLED. Individual dislocations which reached the active ICLED layers exhibited a multiplying effect throughout the structure. Another growth artifact seen in these images was a slow-varying oscillation in the ICLED layers. Our presentation will provide a detailed explanation for both mechanisms and a comparison of the ICLEDs grown on Silicon to those grown on GaSb. Possible strategies for improving the epitaxial quality and device performance will also be discussed.

[1] C. S. Kim et al., Opt. Engr. 57, 011002 (2018).

[2] N. Schäfer et al., Opt. Engr. 58, 117106 (2019).

Modern silicon photonic platforms promise to drastically improve bandwidth, data rate, and power consumption in data centers while data usage is rising rapidly every year. Silicon’s indirect band gap makes it impractical for use in the lasers necessary for transmitting data, requiring integration with other materials such as III-V semiconductors. Monolithic integration by growth of III-V semiconductors on silicon is a promising pathway for cost-efficient production due to the large wafer sizes and ease of packaging and testing compared to hybrid approaches utilizing separate chips [1]. However, on-chip coupling between silicon photonic waveguides and directly grown III-V lasers shows extremely high insertion losses of 7.35 dB [2], significantly reducing the chip’s overall performance.

After etching a window in a patterned silicon photonics wafer, a laser stack can be grown in the opening by molecular beam epitaxy to create a chip where silicon nitride-based waveguides are aligned with InAs quantum dot active regions. Careful calibration of etch and growth rates allows good control of the vertical alignment of the active region and waveguides, but typical MBE growth conditions result in a large horizontal gap between them. While high-performance lasers have been demonstrated on 300mm silicon photonics wafers [3], this gap makes coupling challenging and needs to be addressed for monolithic integration to be competitive with hybrid and heterogeneous silicon photonic implementations.

To improve the coupling between III-V lasers and silicon photonic waveguides, growth conditions must be tailored to minimize the gap between the laser’s active region and the waveguide. Polycrystalline III-V material formed on the silicon dioxide sidewall during the growth of the GaAs buffer causes the crystalline laser stack material to facet and grow away from the sidewall. By increasing the growth temperature, decreasing the growth rate, and decreasing the arsenic overpressure during the growth of the GaAs buffer, the formation of polycrystalline material is suppressed and the horizontal gap between the active region and the waveguide can be reduced from ~5 µm to ~1 µm. Active-passive coupling tests are in-progress and will be reported at the conference.

[1] Z. Zhou et al., “Prospects and applications of on-chip lasers,” eLight 2023.

[2] W.-Q. Wei et al., “Monolithic integration of embedded III-V lasers on Soi,” Light: Science & Applications 2023.

[3] C. Shang et al., “Electrically pumped quantum-dot lasers grown on 300 mm patterned si photonic wafers,” Light: Science & Applications 2022.

The push for scalable silicon photonics drives interest in monolithic integration of InAs quantum dot lasers on the same chip as passive waveguides. One integration method embeds silicon nitride waveguides in SiO₂ across a Si (001) wafer. The SiO₂ is patterned to form rectangular “pockets” where the III-V laser stack is grown. Pocket lasers were recently developed on 300 mm Si (001) wafers, but they have limited optical performance and thermal tolerance compared to devices grown on unpatterned “blanket” Si (001) substrate [1]. Identifying the source of this discrepancy is essential to produce blanket-level device performance within the monolithically integrated platform. One fundamental aspect is the material quality of the pocket-grown III-V epi.

Threading dislocations (TD) generated from the mismatch in III-V/Si lattice and coefficient of thermal expansion diminish laser performance and reliability. In blanket III-V on Si growth, thermal cyclic annealing (TCA) and insertion of dislocation filter layers (DFL) lowers TD density (TDD) by over two orders of magnitude compared to untreated growths [2]. In fully pocket-grown III-V epi on Si, it is impractical to apply TCA due to temperature control and uniformity challenges across a wafer surface containing mixed SiO₂, polycrystalline III-V, and epitaxial III-V regions. Pocket-grown lasers lacking TCA have TDD on the order of 10⁷ cm^-2, suggesting TDD as a major contribution to limited performance compared to blanket-grown lasers.

Here, an annealed GaAs interlayer is inserted between the Si substrate and the patterned SiO₂ to reduce the TDD of in-pocket devices. A planar GaAs buffer is grown on a blanket GaP/Si (001) wafer and exposed to TCA before SiO₂ deposition. After SiO₂ and waveguide patterning, DFLs and the QD laser are grown inside the pockets. The modified structure retains a patterned SiO₂ device configuration while successfully reducing TD presence in pocket-grown material to mid-10⁶ cm^-2. Lasers with continuous-wave (CW) power of 15 mW and CW lasing up to 55 °C are demonstrated and compared to equivalent lasers grown on native GaAs substrates, showing the effect of TDs. The reduced TDD platform helps separate the known performance impact of TDD presence from subtler pocket-specific thermal or geometric effects that may also differentiate the characteristics of pocket and blanket lasers. Thus, these reduced TDD pocket-grown devices lead one step closer to realizing blanket-quality lasers for monolithic integration.

[1] C. Shang, K. Feng, E. T. Hughes, et al. Light Sci Appl 11, 299 (2022).

[2] C. Shang, Y. Wan, J. Selvidge, et al. ACS Photonics 8, 2555−2566 (2021).

Night-vision goggles (NVGs) based on GaAs photocathodes and vacuum-based charge multipliers are highly sensitive with a long battery life exceeding 20 hours [1]. However, the weight and length of the bulk lenses, high-voltage transformer, and fiber bundle inverter lead to neck strain and fatigue for users. Furthermore, the response of the GaAs photocathodes cuts off at 870 nm and cannot fully use the night glow spectrum extending to 1.7 µm. The DARPA ENVISION program aims to create night vision systems that combine thin and lightweight meta-lenses with a solid-state NIR upconverter to reduce the neck torque from current NVGs. In this work, we describe MBE growth of InP/InGaAs npn heterojunction phototransistors (HPTs) and AlInGaP/GaAs visible LEDs, enabling us to demonstrate upconversion of 1.55 µm light to 625 nm visible light. Rapid thermal annealing (RTA) of the LED greatly reduced the forward current needed for visible electroluminescence and promises to increase the upconversion sensitivity by ~6x. The high sensitivity and low power consumption of the devices demonstrated here are promising for next-generation night vision systems.

Multiband photodetectors and FPAs have been developed for various commercial and defense applications such as resources survey, chemical sensing, target seeking, and eye-safe imaging for autonomous automobiles. When implementing multiband photodetectors into an FPA with more than two bands, additional terminals for each pixel greatly complicate the FPA layout and device processing, decrease the fill factor, and increase the ROIC complexity.[1] It is therefore highly desirable to minimize the number of terminals so that the FPA can be integrated with off-the-shelf single-band ROICs. This talk reports the demonstration of multiband monolithically integrated optically-addressed photodetectors using GaSb and InAs/InAsSb type-II superlattices (T2SLs) to cover SWIR, MWIR, and LWIR detection range. The operating principle of the optical-addressing design is to use multiple optical biases on a stack of photodiodes (PDs) connected in series to switch detection bands. The detecting PD is the current-limiting device and determines the spectral response.